You are on page 1of 9

Luminescence properties of CdSe quantum crystallites: interior and surface localized states

M. G. Bawendi, P. J. Carroll, William L. Wilson, and L. E. Brus


AT&TBeN Laboratories, Murray Hill, New Jersey 07974

Resonance

between

(Received 20 August 1991; accepted 1 October 1991) We use time-, wavelength-, temperature-, polarization-resolved luminescence to elucidate the nature of the absorbing and band edge luminescing states in 32 A diameter wurtzite CdSe quantum crystallites. Time-resolved emission following picosecond size-selective resonant excitation of the lowest excited state shows two components-a temperature insensitive 100 ps component and a microsecond, temperature sensitive component. The emission spectrum, showing optic phonon vibrational structure, develops a - 70 wave number red shift as the fast component decays. Photoselection shows the slow component to be reverse polarized at 10 K, indicating this component to be the result of a hole radiationless transition. The 100 ps emitting state is repopulated thermally as temperature increases from 10 to 50 K. All available data are interpreted by postulating strong resonant mixing between a standing wave molecular orbital delocalized inside the crystallite and intrinsic surface Se lone pair states. The apparent exciton transition is assigned to a - 130 wave number wide band of eigenstates with the hole localized principally on the surface. The band contains strongly emitting doorway states and weakly emitting background states. The hole becomes mobile among these states as T increases to 50 K. It is suggested that such resonant mixing may be general in II-VI and III-V crystallites.

I. INTRODUCTION Semiconductor crystallites small with respect to the bulk exciton Bohr radius exhibit quantum confinement in all three dimensions. Their spectroscopic and photophysical properties are essentially those of large molecules. Their excited electronic states occur at higher energy than the bulk band gap and in principle are discrete rather than continuous. Optical studies of nanometer size II-VI quantum crystallites (QCs) have shown that substantial homogeneous widths are present in both absorption and fluorescence. These studies are further complicated by inhomogeneous broadening due to distributions in size, structure, and photophysical dynamics. -l4 Sources of the homogeneous broadening and even the identity of the luminescing state have remained uncertain. Time-resolved temperature-dependent luminescence experiments reported by Eychmuller et al. 3 and O Neil et al. have begun addressing the nature of the luminescing state in CdS QCs and implicate shallow traps as dominating the relaxation dynamics. We have recently found an organometallic method for synthesizing CdSe QCs which are highly luminescent at the band edge.15 A detailed structural study shows near wurJzite crystallites with a narrow size distribution at -32 A. Luminescence excitation and hole burning data reported in a recent letter show a number of resolved electronic states in the homogeneous spectra and suggest that the apparent mirror image intense luminescence of the lowest excited state actually occurs from a surface localized state produced by radiationless transitions on a 160 fs time scale.r6 As these CdSe crystallites are well characterized and exhibit a wealth of spectroscopic detail, we attempt now to understand the nature of electronic excited states in these large molecules. In this paper, we report a detailed time-, wavelength-, temperature-, and polarization-resolved luminescence study.
946 J. Chem. Phys. 96 (2), 15 January 1992

We then consider the absorbing and luminescing states and their dynamics in view of all the spectroscopic data. II. SAMPLE PREPARATION AND EXPERIMENTAL APPARATUS As the data depend critically on sample quality, we describe our procedures in some detail. The QCs were prepared in a two step process which effectively separates the nucleation step from the growth step and which yields unusually good crystalline quality and size monodispersity. CdSe seeds 10-15 A in diameter, absorbing near 400 nm, are made by the micelle method where the water to soap ratio is kept to a minimum by drying the soap and not adding any excess water. 8 A bright yellow 7* powder is recovered and redissolved under argon in a solution of Bu, P (90%) and Bu, PO ( 10% ). The filtered solution is heated under argon for 1 to 3 h at about 200 C. Growth as judged by color change is slow during the first half hour and then more rapid to yield a bright orange-red solution. At this point, further growth is extremely slow, suggesting that there is a bottleneck in the growth process under these conditions. Transmission electron microscopy (TEM) examination shows crystalline particles with an average diameter of 32 A and a standard deviation of < 8% limited by our ability to measure the particle diameters precisely. The apparent growth bottleneck yields a high degree of monodispersity and the high temperature contributes to good crystalline quality. An x-ray powder diffraction study shows wurtzite structure with one stacking fault per crystallite on average.15 Optical measurements were made directly in the growth solutions under argon with small amounts of o-terphenyl added to make an organic glass at low temperature in a Helitrans cryostat. The solution is held between sapphire flats with Teflon spacers and a temperature sensing diode is at@ 1992 American Institute of Physics

0021-9606/92/020946-09$06.00

Downloaded 01 Dec 2006 to 128.111.125.82. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

Bawendi et a/.: Luminescence of CdSe quantum crystallites

947

tached directly to the sapphire window. Luminescence decays were recorded from a partially scattering 1 mm solution. Polarization studies were performed with a 0.1 mm spacer and optically clear untracked regions of 2-3 mm could be obtained upon cooling to 10 K. In these studies, the laser pulses were vertically polarized and both vertically and horizontally polarized luminescence decays were separately recorded. A polarization scrambler was inserted in the collection optics before the spectrometer. A time correlated photon counting apparatus with instrumental resolution of 80-100 ps [full width at half-maximum (FWHM) ] was used to measure the luminescence dynamics. The third harmonic of an Antares Nd:YAG mode locked laser pumped a tunable dye laser to yield 5-8 ps pulses in the 480-570 region. The luminescence was dispersed with a Sopra ultrahigh resolution subtracjive double spectrometer with a typical bandwidth of 0.5 A. In other experiments, time gated spectra were obtained with an intensified OMA (20 ns gate) and a 10 Hz Q-switched Nd:YAG/ dye laser system. Ill. OBSERVATIONS AND ANALYSIS A. Luminescence kinetics Our previous letter describes how photoluminescence excitation and subpicosecond hole burning spectra can be used to unravel and model the homogeneous and inhomogeneous contributions to the ls,--Is,, absorption transition shown in Fig. 1 (a). The extracted homogeneous absorption shows a Franck-Condon progression of three longitudinal optic (LO) vibronic lines spaced 200 cm - apait, as shown in Fig. 3 of Ref. 16. Each vibronic line at low temperature has a homogeneous width of - 130 cm - which could in princi, ple be due to an electronic dynamical process (lifetime broadening ) , or to an unresolved vibrational progression of a low frequency (perhaps acoustic) mode. This structure is obscured in averaging over the inhomogeneous size distribution to give the total absorption spectrum [Fig. 1 (a) 1. Excitation in the bluegives a symmetric band edge emission with high quantum yield (estimated > 0.1) in Fig. 1 (a). The entire size distribution contributes to this emission. In contrast, only the largest crystallites are excited selectively on their zero phonon line by spectrally narrow excitation on the red edge of the absorption spectrum. Figure 1 (b) shows the resulting cw structured emission. This emission approaches that of a perfectly monodisperse sample. There is a Franck-Condon LO phonon emission progression with bulk spacing of 200 cm - with a shift of - 75 , cm - between the excitation energy and the peak of the LO zero phonon line in emission. This line has a 75 cm - mostly inhomogeneous width. As we found previously, the homogeneous width in emission is only < 20 cm - in contrast to the , 130 cm - width of this same LO zero phonon line in absorption.lb (The excitation spectrum of this structured emission yields structured absorption containing an LO phonon progression and higher excited states, as shown in our letter.) The shape of the structured emission spectrum is nearly insensitive to excitation wavelengths on the red side of the

Luminescence

Absorption

91 0

Pum

450

500
Wavelength (nm)

550

10

FIG. 1. (Top) Absorption and luminescence spectra of 32 A CdSe crystallites at 15 K. The excitation light was at 440 nm. The 15 nm shift between absorption and luminescence is a characteristic of the distribution in particle sizes and not a single particle property as explained in the text. A small amount of deep red emission -600-700 nm is not shown. (Bottom) The same absorption spectrum, but the luminescence excitation wavelength is on the red edge of the absorption band (550 nm) for size selection. This leads to significant line narrowing and the appearance of an LO phonon progression. The shift between the excitation and the peak of the LO zero phonon is 75 cm- .

absorption, but as bluer light is used, the apparent ratio of the intensities of the zero to first LO phonon decreases until the first LO phonon appears more intense than the zero LO phonon. This is a direct result of the size distribution-bluer wavelengths excite a broader distribution by exciting large crystallites on their first LO phonon and smaller ones on their zero phonon line. The apparent first LO phonon line is then a sum of a zero LO phonon line from larger particles and a first LO phonon line from smaller particles. The - 15 nm shift between absorption and emission in Fig. 1 (a) is a characteristic of the distribution and not of single QCs. Time-resolved emission spectra constructed from decay curves at different emission wavelengths appear in Fig. 2. This data was taken with - 1 A bandpass at 15 K. There is a clear red shift of the structured spectrum during the first nanosecond. The 75 cm- shift between excitation wavelength and the peak of the LO zero phonon emission line appears on the time scale of hundreds of picoseconds. In Fig. 2, the highest intensity at each time is scaled to the same absolute value; there is no resolved rise time in the emission at any wavelength in this region. The spectrum stops sliding red after 1 ns and the shape is identical to that of the cw spectrum. Spectral evolution on this time scale is not present in our previous pump-probe absorption experiment, where

J. Chem. Phys., Vol. 96, No. 2,15 January 1992 Downloaded 01 Dec 2006 to 128.111.125.82. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

948

Bawendi eta/.: Luminescence of CdSe quantum crystallites Time Resolved Luminescence Luminescence Decays Log Absolute Intensities) IoK

Time (nsec)
FIG. 4. Luminescence decays of the LO zero phonon line during the first 2 ps as a function of temperature.

550

555

565 560 Wavelength (nm)

570

FIG. 2. Time-resolved luminescence spectra showing a red slide (75 c m - ) ofthe spectrum as a function of time during the first nanosecond. The intensities are all scaled to the same value for comparison of line shapes. The resolution of the apparatus was 8C-100 ps and 0.7 A. The excitation was at 549 nm.

the ground state bleach did not evolve from 0.5 ps to several nanoseconds. 6 Figures 3 and 4 show the emission decay at the peak of the zero LO phonon line as a function of temperature plotted using absolute emission intensities. There are two distinct components-a short, temperature insensitive component

on the 100 ps scale and a long multiexponential microsecond component whose average lifetime decreases by a factor of about - 10 from 10 to 50 K. Because absolute intensities measured under identical conditions are used, the small decrease in signal at short times and the increase in the intercept of the long component as the temperature increases in Fig. 3 are meaningful. Similarly, the decrease in integrated area with increasing temperature in Fig. 4 reflects a drop in the quantum yield. Emission data taken further red, on top of the higher LO phonons show little quantitative change, as would be expected from Fig. 2. Figure 5 shows a two exponential decay fit and the instrumental impulse response. The fit yields a lifetime of 110 ps for the short component. Figure 6 shows the temperature dependence of the cw emission plotted using absolute intensities. There are two luminescence bands-the band edge emission discussed above and deep trap emission in the near IR. The relative

LUMINESCENCE DECAY

Luminescence Decays (Absolute Intensities) 30K 20K 10K INSTRUMENT RESPONSE

I 1

I 2

I 3

I 4

I 5

I 6 0 2 4 TIME ( nsec ) 6 8

lime (nsec)
FIG. 3. Luminescence decays of the LO zero phonon line during the first 6 ns as a function of temperature.

FIG. 5. Fit of the luminescence decay to two exponentials and instrument response. The fast decay is fit with a lifetime of 110 ps.

J. Chem. Phys., Vol. 96, No. 2,15 January 1992


Downloaded 01 Dec 2006 to 128.111.125.82. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

Bawendi eta/.: Luminescence of CdSe quantum crystallites

949

lb>
C W LUMINESCENCE Ic>

440

520

600 680 WAVELENGTH ( nm )

760

la>
FIG. 8. The three-state model. yti is the radiative decay from b to a, yk is the radiationless decay from b to c, and yrb represents the rate of repopulation of 6 fixed by microscopic reversibility. There is an energy offset E between band c. We have put in a nonradiative pathway rnr to account for the decrease in quantum yield with temperature. We also assume b and c have equal degeneracies.

FIG. 6. Steady state luminescence as a function of temperature. There is a decrease in the band edge luminescence quantum yield as the temperature is increased. The deep trap emission, however, increases with temperature.

quantum yield of band edge emission decreases by a factor of 3 from 10 to 50 K, in the same range where the long component lifetime decreases by a factor of - 10. The deep trap emission increases with temperature. Despite the fact that the figure appears to show an isobestic point, these two emissions originate in separate groups of QCs. The excitation spectrum of the trap emission is blue shifted from that of the edge emission and the deep trap emission appears without resolvable ( < 500 ps) rise time. A rise time would be expected if the edge emission were feeding the deep trap luminescence. Figure 7 shows the temperature dependence of the LO phonon structured edge emission. This data was taken on the nanosecond laser apparatus with a 20 ns gate. As temperature increases, the spectrum shifts to higher energy and emission occurs to the blue of the exciting wavelength, consistent

n
I
5k

GatedLuminescence

Pump

5io

565

560

Wavelength (nm)
FIG. 7. Temperature dependence of the luminescence spectrum. The shift between absorption and the zero phonon peak in luminescence decreases with temperature. The pump laser was almost all gated out except for a

with populating states within kT of each other. The LO phonon linewidths also increase with temperature. The temperature dependence of the long component lifetime, relative quantum yield, and apparent initial radiative intensity in Figs. 3-5 indicate that complex dynamics occur for T< 50 K. Figure 2 indicates that both the fast and slow components show the same LO Franck-Condon progression. The fast component is resonance emission, while the slow component is red shifted by about 75 cm - at 10 K. Moreover, in the next section it is shown that the long component is not directly excited by the laser, but is produced by radiationless transition from some other initially excited state. In Figs. 3 and 4, the changes with temperature are suggestive of fast thermal equilibrium between a weakly emitting, long-lived lower state, and a strongly emitting, upper state. (A thermal repopulation effect in band edge emission has been observed previously in two recent studies.9* ) 3 These results can be modeled phenomenologically by assuming that both fast and slow components occur in one QC and do not represent emission from separate groups of QCs. Figure 8 outlines schematically a three level model of this sort. (a) is the ground state, (b) represents initially populated states which carry most of the oscillator strength in absorption, while (c) represents darker states populated principally by radiationless transitions from (b) . The definition of the rates and the final fit is given in the captions of Figs. 8 and 9. We have fit this model to our data. The thermally derived energy difference (E = 50 cm - ) between (c) and (b) matches the spectroscopic shift between fast and slow components (75 cm - ) fairly well. The derived fit parameters are well determined with each one fixed by a different part of the data. The experimental decays in Figs. 3-4 are well reproduced by the three level model fit of Figs. 9-10, where the instrumental response function has been convoluted with the calculated decays. The multiexponential character of the experimental long decay can be attributed to a distribution of rates among the particles. Also note that we earlier concluded from analysis of the coher-

smallamountwhichwaskept in for calibration.

ent couplingtransientin the bleachspectrumon the 300fs


J. Chem. Phys., Vol. 96, No. 2,155January 1992

Downloaded 01 Dec 2006 to 128.111.125.82. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

950

Bawendi et&:

Luminescence of CdSe quantum crystallites

h
x .e 2 s +

Calculated Luminescence

transitions of excited states through the intensity and kinetics of luminescence polarization. The alignment (1) r(t) = (I,, -I,,/U,, + 21,) measures the magnitude of emitting dipole alignment independent of population decay.* A positive (negative) r(t) implies that the emitting dipole is principally parallel (perpendicular) to the initial absorbing dipole. K(t) measures population decay independent of changes in emitting dipole alignment K(t)+ + 21*. (2) Figure 11 (a) shows that the 10 K luminescence at the LO zero phonon emission peak is initially polarized parallel with r = 0.1 and then changes to r = - 0.095 in the slow microsecond decay. The fact that the slow component is negatively aligned with respect to the initial absorbing dipole unambiguously implies that the slow component is not directly excited by the laser, but instead created principally by radiationless transition. Since r( 1) changes on the same time scale as the photon counting instrumental response function P(t), we must deconvolute our data. We form an analytic fit to K(t) for the initial 500 ps by convolution to the measured P(t). As de-

30K %
I 1 I I 1 I

lime (nsec)
FIG. 9. Calculated decays during the first 6 ns using E = 75 K, yaa = 0.1 , , ns- ym = 0.0045 ns- ybE= 10 ns- and ynr (IO, 20, 30, and 40 , K) = (0,0.0025,0.007, and 0.015). The instrumental response function in Fig. 5 has been convoluted with the calculated decays. yrb is determined by microscopic reversibility. The QYwas assumed unity at 10 K. yk is almost uniquely determined by the initial fast decay and yCSis almost uniquely determined by the long decay at 10 K, E is determined by the ratios of the intercepts for the long decays as a function of temperature and ya6 is determined by the ratios of the intercepts for the long and short decays. ynr were chosen to fit the decrease in QY for the long decays as a function of temperature. An almost identical fit can be obtained if the QYis assumed to be 0.1 at 10 K by taking yob = 0.01, y- = O.ooO45, ynr( 10, 20, 30, and 40 K) = (0.093,0.067,0.056, and 0.044), and Eand ybeas before. In this case, the long lifetime is dominated by nonradiative decay at 10 K rather than radiative decay. QY is the luminescence quantum yield.

scale that these QCs had to be characterized by a three level system with some sort of long lived trap state.16 6. Polarization alignment kinetics Even though the QC lattice axes have no alignment or orientation in laboratory axes, polarized optical excitation can create an aligned excited state due to the angular dependence of the transition dipole matrix element. This process, termed photoselection in molecular spectroscopy, provides information on the symmetry and radiationless

TIME RESOLVED LUMINESCENCE POLARIZATION

200

400 TIME ( psec )

600

800

4 x .?i E $2 5 T 0
I I I 1

0.05 F 5 iii 0.00 --____---_____---__------

r------.+---

10K 20K 30K 4OK

5 d -0.05 ;;
l5 f -0.10 : a
-0.15 0
&A AA A

I 10

500

1000

1500

2000

I I I 20 30 40 TEMPERATURE ( K )

50

60

Time (nsec)
FIG. 10. Calculated decays for the first microseconds using the parameters of Fig. 9.

FIG. 11. (Top) Luminescence polarization alignment decays. III and Zr are the parallel and perpendicular components of the luminescence. The alignment switches from parallel to perpendicular during the first 100 ps. (Bottom) Temperature dependence of the polarization alignment.

J. Chem. Phys., Vol. 96, No. 2,15 January 1992


Downloaded 01 Dec 2006 to 128.111.125.82. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

Bawendi et&:

Luminescence of CdSe quantum crystallites

951

scribed above, we assume a two state, sequential decay model for r(t), where a directly excited state b emits with alignment R, and lifetime I-. The final, slowly decaying state c has an infinite lifetime on the 500 ps scale and a directly measured r [ - 0.095 in Fig. 11 (a) ] r(t) = [ (R, + 0.095)e - - 0.0951. R, and I- are determined by fitting P(t)K(t)r(t)dt (4) I for both absolute magnitude and shape. In this fit, the two parameters are coupled with an increase of one compensating for a decrease in the other. There are two important physical possibilities for R,-O.15 for an excitation dipole degenerate in a plane and 0.4 for a linear dipole. Figure 12 (a) shows that an initial 0.15 alignment fits the data well if r = 65 ps. A 0.4 alignment does not fit as the alignment peaks too early and decays too rapidly. We conclude from this fit that the initial alignment is low, -0.15, and decays on about the same time scale as the fast b state in the population decay. Because of the coupling between R, and r, we cannot determine whether the two lifetimes are actually exactly the same.
TWO STATE MODEL LUMINESCENCE POLARIZATION

(3)

I,, -1,

Figure 11 (b) shows that the slow component alignment is very sensitive to temperature and decreases to near zero (isotropic emission) at 25 K. Figure 12(b) shows the temperature dependence of the deconvoluted r assuming that R, = 0.15 independent of temperature. Near 50 K, r decreases by about a factor of 2. In order to understand the variation of alignment across the inhomogeneous distribution at 10 K and to search for changes in initial dipole direction with excitation wavelength, we used exciting wavelengths within a 900 cm - region centered at 545 nm. At each wavelength, we observed luminescence on the zero phonon line shifted 70 cm - from excitation and on the vibronic bands shifted 270 and 470 cm - The alignment and decay time of the fast component . are about the same under all circumstances. For each excitation wavelength, the alignment in the slow component is negative in the range - 0.05 to - 0.12 on the zero phono line, but closer to zero on the vibronic members shifted further red. The intensity of the slow component is typically one-third higher when observed on the red shifted vibronic members. Additionally, it was observed in a sample irradiated repeatedly over three months that positively aligned slow component states, decaying on a 5 ns time scale, grew in with time. In this photolyzed sample, the alignment switched from positive to negative after - 10 ns. (These photoproduct states were not observed in samples stored under argon for six months in room light. ) IV. DISCUSSION AND ASSIGNMENT A. Internal effective mass QC electronic states Effective mass models consider crystallite internal molecular orbitals ( MOs) that evolve into the continuous valence and conduction bands in the bulk crystal. This model ignores possible surface states and surface reconstruction. The MOs are standing waves with surface nodes formed from Bloch eigenfunctions. It is important to consider the chemical nature of the MOs. The lowest transition is ls,-lsh, indicating that the standing wave envelope function is the same for both electron and hole. However, the basis wave functions inside the unit cell are different-the electron basis is essentially an s orbital on Cd, while the hole basis is a threefold degenerate p orbital on Se.* The direction of thep orbital with respect to crystallite axes gives the polarization of the transition dipole. Within the unit cell, the valence to conduction band electronic transition is threefold electronically degenerate and isotropic. Spin-orbit coupling is strong in Se and as a result the highest occupied hole orbital Is, transforms like a F = 3/2 (F = L + J) state with fourfold degeneracy.22*23 aspheriIn cal zinc-blende crystallite, the lowest transition involving this fourfold degenerate state would be isotropic and photoselection would not occur. Our QCs have a uniaxial near wurtzite structure. Wurtzite can be considered as a weak perturbation on the Td zincblende structure and the degenerate lowest ls,-lsh transition splits into a lower F, = 3/2 A exciton and an upper

-----_---_--------

-----.

(a)

200

400
TIME ( psec )

600

800

0
(b)

IO

20

30

40
(K)

50

60

TEMPERATURE

FIG. 12. (a) (Top) Polarization alignment of the luminescence fit to a two upper state model as in Fig. 8. The parameters are described in the text. (b) (Bottom) Temperature dependence of the fit fast polarization decay ~for a

fixedR, = 0.15.

F, = l/2 B exciton(wheretheA4 notationis borrowed


J. Chem. Phys., Vol. 96, No. 2,15 January 1992

Downloaded 01 Dec 2006 to 128.111.125.82. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

952

Bawendi eta/: Luminescence of CdSe quantum crystallites

from the bulk). 24 The A-B splitting is 204 cm- in the bulk2 and is predicted to be one-sixth as large in QCS.~~ Both excitons have radiative lifetimes on the order of 1 ns or less. In both A and B excitons, the hole effective mass is quite anisotropic. in A, the hole is principally on Se px and p,, orbitals in the sheet perpendicular to the unique wurtzite z axis; as the electron is in an s orbital on Cd, the A transition dipole is polarized principally in these planes. In the B exciton, the hole is inp, and the transition dipole is principally along z. This A-B wurtzite internal MO structure cannot explain our data. We do have a two level system with different transition dipole directions, but the lower state has an extremely long, essentially forbidden radiative lifetime Also, if the B exciton were excited and relaxed quickly to give short component A emission, then the short component polarization would have reversed as a function of increasing excitation-emission splitting. B to A relaxation occurs in about 1 ps in bulk wurtzite CdSe.* B. Surface electronic states

C. Model for QC states and dynamics State c has an extremely long lifetime implying that at least one carrier is localized or trapped, producing a very poor overlap with the other carrier. Moreover, as we have seen, the transition dipole direction is determined by the direction ofthep orbital in which the hole resides. The fact that the c component shows reverse polarization at 10 K implies that the hole must somehow participate in the b to c radiationless transition. Thus we propose the hole is surface localized, essentially at the same energy as the internal Is, MO, as suggested by the surface band calculations, and as conjectured in our letter. We suggest the electron remains in Is,. The homogeneous spectroscopic data in our previous letter, as we11as our present kinetic and thermal data, can be understood if there is strong resonant mixing among the zero-order A and B internal MOs and the zero-order surface Se lone pair states. There are about 130 surface Se atoms. Some will couple strongly and some will couple weakly to the interior A and B MOs; the situation is similar to intermediate strong coupling in molecular radiationless transition theory.32 The left-hand side of Fig. 13 shows the unmixed states where Is,, is shown by a long line to indicate that it carries virtually all the oscillator strength in transitions from the ground state. The band of states depicted by short lines is the narrow band of surface Se lone pair states. These states are virtually dark. After mixing on the right-hand side, there are now a modest number of doorway states that for reasons of geometry and energy carry most of the oscillator strength, and a large number of background states which have little Is, character. The width of the distribution is about 130 cm- , explaining the width of the LO zero phonon line in the homogeneous absorption spectra.16 We assume that with one electron-hole pair in the crystallite, the electronic structure is now sufficiently different that all right-hand side eigenstates are removed from the absorption spectrum. Optical excitation on the picosecond time scale directly creates

As we have seen, each 32 A diameter QC has an internal F = 3/2 highest occupied molecular orbital (HOMO). This state lies about 0.1 eV below the top of the bulk valence band. The QC has about 800 atoms, one-third of which are on the surface. Are there any surface electronic states at similar energies? We envision a compact faceted QC and turn for guidance to recent studies of nonpolar wurtzite CdSe cleavage surfaces. Theory predicts a surface HOMO band of narrow width (heavy effective mass), composed of lone pairs on Se atoms that have three covalent bonds into the bulk lattice (labeled S, in Ref. 28). This surface band actually lies within the band gap if the surface geometry is held unchanged from bulk tetrahedral angles and bond lengths. That is, an internal HOMO hole could localize spontaneously on the surface. Because there is a large increase in effective mass from a delocalized hole to one localized in the surface band, the kinetic energy of the hole is not increased substantially upon localization, as would be expected from bulk arguments alone. Surface reconstruction, confirmed by experiment, involves bond angular rotation that brings the three Se to Cd bonds closer to pyramidal geometry.29*30This rotation stabilizes the lone pair surface band, bringing it just below (0.2 eV) the valence band top. There is also a surface localized rotational phonon mode due to this reconstruction, predicted to occur at 40 cm - in CdSe.3 [The existence of such a low frequency surface localized mode is confirmed experi mentally for a similar case in GaAs ( 110). ] The hole stabilization energy in one lone pair is sensitive to motion in this mode. There is an equivalent surface lowest-occupied molecular orbital (LUMO) band composed of empty s orbitals on surface Cd atoms about 0.5 eV above the conduction band minimum. In our Lewis base phosphine solvent, strong dative bonds to surface Cd atoms will push these empty orbitals to higher energy.

Surface hole traps

mixed

states

FIG. 13. The model for the electronic structureof the valence band. (left) The single Is,, state which carries almost all the oscillator strength mixes with a large number of localized surface states which are nearly dark. Some surface states mix more strongly than others as explained in the text. This results in doorway states with moderate oscillator strength and darker background states which do not carry much oscillator strength (right). The electronic states are probably sufficiently broadened by coupling to low frequency modes that they overlap to form an effective continuum.

J. Chem. Phys., Vol. 96, No. 2,15 January 1992


Downloaded 01 Dec 2006 to 128.111.125.82. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

Bawendi etal.: Luminescence of CdSe quantum crystallites

953

doorway b states that emit resonantly with a radiative lifetime > 10 ns, too long for b to be a pure A or B exciton. The principle decay channel of 6, however, is fast radiationless transition via hole hopping on the surface into the background states c. As temperature increases, the hole is mobile on the surface and thermally repopulates higher energy b and c states. The initial fast component shows alignment suggestive of the A exciton. The zero-order A internal MO is delocalized in the planes perpendicular to z. At the edges of these sheets, there are Se atoms with two bonds into the sheet and one bond to a neighboring sheet. These are the same Se atoms involved in the S, surface state. These atoms should couple strongly to A and doorway luminescence from them would retain the A polarization. We do not have a physical picture why the lowest energy background states should be polarized perpendicularly. It seems reasonable that as temperature is raised, hole motion among background states samples all types of lone pair alignments and produces isotropic emission. The 40 cm - rotational phonon would be effective in promoting surface motion, in qualitative agreement with the kinetics observed at < 50 K. Relaxation along this mode may also be instrumental in initially localizing the hole on the surface. On our 32 A diameter QC, we must have many physically inequivalent Se atoms, unlike the flat surfaces modeled theoretically. On the QC, we probably have a disordered, narrow surface band S, with the lowest states localized and higher states partially delocalized on the surface. This idea that the electronic wave function of both b and c luminescing states is fundamentally different from that of the absorbing state also rationalizes two facts-the LO Franck-Condon pattern is different in emission from that in absorption and the homogeneous width of relaxed emission LO lines is quite narrow as compared with the same width in absorption. An alternate interpretation of the 130 cm - width in absorption and red shift of the luminescence might involve acoustical phonon coupling to unperturbed A excitons. The width would be a Franck-Condon contour and the picosecond red shift of the luminescence would represent slow phonon relaxation. However, this hypothesis does not explain the reversed polarization, extremely long radiative rate, or narrow homogeneous width of LO vibronic lines in the c state. There may be some coupling to acoustical phonons, but it is not the dominant effect. D. General comments and speculations

Nevertheless, this strong resonance may be common. Surface reconstruction on flat surfaces of II-VI and III-V semiconductors in general drives surface states out of the bulk band gap and into resonance with the bulk bands.35 (The optically strong surface HOMO-LUMO transitions appear to occur at quite high energy and this is why the lower energy, low resolution QC absorption spectrum can be understood using only effective mass states.) Theoretical studies for II-VI and III-V materials generally predict the surface HOMO band to be near the top of the bulk valence band and localized on the more electronegative eIement.340 To our knowledge, no experimental study of broadening in QCs has ever shown a narrow homogeneous width in absorption; resonance broadening may occur in many of these cases, especially in the tight confinement limit. Larger QCs, with diameter approaching the bulk Bohr exciton diameter, may show less resonant mixing and perhaps genuine exciton Iuminescence. Besides luminescence, such resonance will be important in the quantum mechanics of charge transfer across the interface. As mentioned previously, two recent studies of CdS QC band edge luminescence have also observed complex, somewhat similar emission kinetics and have modeled exciton thermal repopulation by shallow traps of long radiative lifetime. The careful and thorough works of Eychmuller et aZ.13 and O Neil et aZ.9 postulate a subpicosecond time scale radiationless transition of the electron into surface states spread over a narrow range of energy. This idea is analogous to the strong resonance model we propose here if the role of hole and electron are reversed. It would be interesting to investigate photoselection in these systems. ACKNOWLEDGMENTS We thank M. L. Steigerwald, T. D. Harris, S. W. Koch, and Al. L. Efros for stimulating discussions.
A recent quantum mechanical review is M. G. Bawendi, M. L. Steigerwald, and L. E. Brus, Annu. Rev. Phys. Chem. 41,477 ( 1990). A general review of QC photochemistry is A. Henglein, Top. Current Chem. 143, 113 (1988). *A. P. Alivisatos, A. Harris, N. Levinos, M. Steigerwald, and L. E. Brus, J. Chem. Phys. 89,400l (1988). 3 A. P. Alivisatos, T. D. Harris, P. J. Carroll, M. L. Steigerwald, and L. E. Brus. J. Chem. Phvs. 90. 3463 t 1989). 4P. Roussignol, D. Ricard, C. Flytzanis, and N. Neuroth, Phys. Rev. Lett. 62,312 (1989). 5 D. Ricard, P. Roussignol, F. Hache, and C. Flytzanis. Phvs. Status Solidi B 159,275 (1990). Peyghambarian, B. Fluegel, D. Hulin, A. Migus, M. Joffre, A. AntonN. etti, S. W. Koch, and M. Lindberg, IEEE J. Quantum Electron. 25, 2516 (1989). Z. Hu, S. W. Koch, M. Lindberg, N. Peyghambarian, E. L. Pollack, Y. ,_,. and F. Abraham, Phvs. Rev. Lett. 64, 1805 (199o) Esch, B. Fluegel, G. Khitrova, H. M. Gibbs, X. Jiajin, K. Chang, S. W. V. Koch, L. C. Liu, S. W. Risbud, and N. Peyghambarian, Phys. Rev. B 42, 7450 (1990). M. G Neil, J. Marohn, and G. McLendon, J. Phys. Chem. 94, 4356 (1990). E. Hilinski, P. Lucas, and Y. Wang, J. Chem. Phys. 89,3435 ( 1989). Y Wang A. Suna, J. McHugh, E. Hilinski, P. Lucas, and R. D. Johnson, J. Chem. Phys. 92,6927 (1990). T. Inokuma, T. Arai, and M. Ishikawa, Phys. Rev. B 42, 11093 ( 1990). A Eychmuller, A. Hasserlbarth, L. Katsikas, and H. Weller, Ber. Bun-

It has long been known that QCs can emit from surface localized deep trap states, producing strongly red shifted and broad emission.33*34The dynamics are characteristic of donor-acceptor emission with a distribution of distances. These surface states are thought to be surface vacancies, adatoms, and/or adsorbed foreign species. In this paper, we establish, at least in this one case, that structured QC band edge emission results from quantum mechanical resonance between exciton MOs and intrinsic surface MOs. In a sense, this result is a consequence of an accidental energy

degeneracy the zero-orderstates. of

se&s.Phys.Chem,95,79 (1991).
J. Chem. Phys., Vol. 96, No. 2,15 January 1992

Downloaded 01 Dec 2006 to 128.111.125.82. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

954

Bawendi eta/: Luminescence of CdSe quantum ctystallites


28Y. R. Wang and C. B. Duke, Phys. Rev. B 37,6417 (1988). 29T. N. Horsky, G. R. Bandes, K. F. Canter, C. B. Duke, S. F. Homig, A. Kahn, D. E. Lessor, A. P. Mills, Jr., A. Paton, K. Stevens, and K. Stiles, Phys. Rev. Lett. 62, 1876 (1989). MC. B. Duke, D. E. Lessor, T. N. Horsky, G. Brandes, K. F. Canter, P. H. Lionel, A. P. Mills, Jr., A. Paton, and Y. R. Wang, J. Vat. Sci. Technol. A 7,203O (1989). 3C. B. Duke and Y. R. Wang, J. Sci. Vat. Technol. B 7, 1027 ( 1989). ,*S. Mukamel and J. Jortner,in Excited States, edited by E. Lim (Academic, New York, 1977), Vol. 3, pp. 58-105. ) Chestnoy, T. D. Harris, R. Hull, and L. E. Brus, J. Phys. Chem. 90, N 3;93 (1986). 34A Fojtik, H. Weller, U. Koch, and A. Henglein, Ber. Bunsenges. Phys. Chem. 88,969 ( 1984). H. Carstensen, R. Claessen, R. Manzke, and M. Skibowski, Phys. Rev. B 41,988O (1990). 36C. B. Duke and Y. R. Wang, J. Vat. Sci. Technol. B 6, 1440 ( 1988). C. A. Swarts. T. C. McGill. and W. A. Goddard III, Surf. Sci. 110, 400 (1981). R. Chang and W. A. Goddard III, Surf. Sci. 144,311 (1984). 39S. B. Zhang and M. L. Cohen, Surf. Sci. 172,754 ( 1986). 40R. P. Beres, R. E. Allen, and J. D. Dow, Solid State Commun. 45, 13 (1983).

K Misawa, H. Yao, T. Hayashi, and T. Kobayashi, J. Chem. Phys. 94, 4131 (1991). M G. Bawendi, A. R. Kortan, M. L. Steigerwald, and L. E. Brus, J. &em. Phys. 91,2782 (1989). 16M. G. Bawendi, W. L. Wilson, L. Rothberg, P. J. Carroll, T. M. Jedju, M. L. Steigerwald, and L. E. Brus, Phys. Rev. Lett. 6.5, 1623 (1990). M. L. Steiaerwald. A. P. Alivisatos, J. M. Gibson, T. D. Harris, A. R. Kortan, AYJ. Mulier, A. M. Thayer;T. M. Duncan, D. C. Douglass, and L. E. Brus, J. Am. Chem. Sot. 110,3046 (1988). *A. R. Kortan, R. Hull, R. L. Opila, M. G. Bawendi, M. L. Steigerwald, P. J. Carroll, and L. E. Brus, J. Am. Chem. Sot. 112, 1327 (1990). 19A. C. Albrecht, J. Mol. Spectrosc. 64, 376 (1961). 20For example, see A. J. Cross and G. R. Fleming, Biophys. J. 46, 45 (1984). *W. A. Harrison, Electronic Structure (Freeman, San Francisco, 1980). J. B. Xia, Phys. Rev. B 40,850O (1989). z3G. B. Grigoryan, E. M. Kazaryan, Al. A. Efros, and T. V. Yazeva, Sov. Phys. Solid State 32, 103 1 ( 1990). * J. Hopfield, Phys. Chem. Solids 15, 97 ( 1960). J. 25R. G. Wheeler and J. 0. Dimmock, Phys. Rev. 125.1805 (1962). *6 Al. A. Efros (private communication). D. Braun, W. W. Ruhle, and J. Collet, in Springer Series in Chemical Physics, edited by C. B. Harris, E. P. Ippen, G. A. Mourou, and A. H. Zewail (Springer, Berlin, 1990), Vol. 7, p. 27 1.

J. Chem. Phys., Vol. 96, No. 2,15 January 1992


Downloaded 01 Dec 2006 to 128.111.125.82. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

You might also like